Case Study 1: From Exploration to Design

• Name: Frederick Song
• Student ID: 76798644
• Link to GitHub Repository: DSCI 200 Case Study 1

This blank notebook is provided for you to use to answer questions from Case Study 1.

---

devtools::install_github("lbelzile/hecedsm")

library(tidyverse)
library(hecedsm)

Skipping install of 'hecedsm' from a github remote, the SHA1 (df0a5444) has not changed since last install.
  Use `force = TRUE` to force installation

── Attaching core tidyverse packages ──────────────────────── tidyverse 2.0.0 ──
✔ dplyr     1.1.4     ✔ readr     2.1.6
✔ forcats   1.0.1     ✔ stringr   1.6.0
✔ ggplot2   4.0.1     ✔ tibble    3.3.0
✔ lubridate 1.9.4     ✔ tidyr     1.3.1
✔ purrr     1.2.0     
── Conflicts ────────────────────────────────────────── tidyverse_conflicts() ──
✖ dplyr::filter() masks stats::filter()
✖ dplyr::lag()    masks stats::lag()
ℹ Use the conflicted package (<http://conflicted.r-lib.org/>) to force all conflicts to become errors

Attaching package: ‘hecedsm’


The following object is masked from ‘package:stringr’:

    words

---

Milestone 1: Exploratory Data Analysis (EDA)

Task 1: Choose your favourite dataset

1.

data(BRLS21_EDA)
head(BRLS21_EDA)

A tibble: 6 × 6

id	reltime	task	occurence	eda	hr
<fct>	<int>	<fct>	<int>	<dbl>	<dbl>
1	0	baseline	1	20.60959	88.49560
1	1	baseline	1	21.41062	88.31564
1	2	baseline	1	22.26244	78.13892
1	3	baseline	1	24.22100	65.82009
1	4	baseline	1	24.23851	65.13290
1	5	baseline	1	23.63703	66.83456

---

2.

The primary research questions or objectives the authors aimed to address

• Compare the impact of different methods of sending and receiving messages while driving (Phone, Watch, Speaker, Texting) to assess the level of driving danger.

The importance or relevance of these goals in the context of the field

• While laws prohibit the use of mobile phones while driving, laws do not prohibit the use of smartwatches while driving. However, as they can also be a distraction, they may be a safety risk in driving, which requires sustained attention.
• With smartwatches becoming increasingly common, it is necessary to investigate their impact on driving. Given the limited existing research on this topic, more in-depth studies with larger sample sizes are necessary.

Specific hypotheses or outcomes the study sought to explore

H1 Gaze distribution

• H1a. Gaze distribution is less focused on the driving task while receiving written notifications on a smartwatch than while receiving them on a smartphone.
• H1b. Gaze distribution is less focused on the driving task while receiving written notifications on a smartphone than when hearing them through a speaker.
• H1c. Gaze distribution is less focused on the driving task when drivers have to respond to written notifications by text instead of responding to them vocally.

H2 Driving behavior

• H2a. Receiving written notifications on a smartwatch has more negative consequences on driving behaviour than receiving them on a smartphone.
• H2b. Receiving written notifications on a smartphone has more negative consequences on driving behaviour than hearing them from a speaker.
• H2c. Responding to written notifications by text has more negative consequences on driving behaviour than responding to them vocally.

H3 Mental workload

• H3a. Mental workload is higher while receiving written notifications on a smartwatch than on a smartphone inside a driving simulator.
• H3b. Mental workload is higher while receiving written notifications on a smartphone than when receiving them vocally inside a driving simulator.
• H3c. Mental workload is higher while responding to notifications by text than by responding orally to notifications inside a driving simulator.

---

Task 2: Explore data

glimpse(BRLS21_EDA)

Rows: 10,942
Columns: 6
$ id        <fct> 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, …
$ reltime   <int> 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17…
$ task      <fct> baseline, baseline, baseline, baseline, baseline, baseline, …
$ occurence <int> 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, …
$ eda       <dbl> 20.60959, 21.41062, 22.26244, 24.22100, 24.23851, 23.63703, …
$ hr        <dbl> 88.49560, 88.31564, 78.13892, 65.82009, 65.13290, 66.83456, …

---

3.

The dataset contains 10,942 observations and 6 variables.

---

4.

The types of variables in the dataset are:

• categorical: id and task.
• numerical: reltime, occurence, eda and hr.

---

5.

There is no need to convert any variable in the dataset before starting the analysis.
The current variable type is appropriate and can be used directly for data visualization.

---

6.

BRLS21_EDA |>
  summarise(
    id_missing        = sum(is.na(id)),
    reltime_missing   = sum(is.na(reltime)),
    task_missing      = sum(is.na(task)),
    occurence_missing = sum(is.na(occurence)),
    eda_missing       = sum(is.na(eda)),
    hr_missing        = sum(is.na(hr))
  )

A tibble: 1 × 6

id_missing	reltime_missing	task_missing	occurence_missing	eda_missing	hr_missing
<int>	<int>	<int>	<int>	<int>	<int>
0	0	0	0	214	538

BRLS21_EDA |>
  summarise(
    id_missing_prop        = sum(is.na(id))        / nrow(BRLS21_EDA),
    reltime_missing_prop   = sum(is.na(reltime))   / nrow(BRLS21_EDA),
    task_missing_prop      = sum(is.na(task))      / nrow(BRLS21_EDA),
    occurence_missing_prop = sum(is.na(occurence)) / nrow(BRLS21_EDA),
    eda_missing_prop       = sum(is.na(eda))       / nrow(BRLS21_EDA),
    hr_missing_prop        = sum(is.na(hr))        / nrow(BRLS21_EDA)
  )

A tibble: 1 × 6

id_missing_prop	reltime_missing_prop	task_missing_prop	occurence_missing_prop	eda_missing_prop	hr_missing_prop
<dbl>	<dbl>	<dbl>	<dbl>	<dbl>	<dbl>
0	0	0	0	0.01955767	0.04916834

• The variables id, reltime, task, and occurence have 0 missing values.

• The variable eda has 214 missing values. Its proportion of missing data is 0.01955767.

• The variable hr has 538 missing values. Its proportion of missing data is 0.04916834.

• There are no variables with a large proportion of missing data.

---

7.

BRLS21_EDA |> count(task) |> mutate(prop = n / sum(n)) |> arrange(-prop) |> head(4)
BRLS21_EDA |> count(id)   |> mutate(prop = n / sum(n)) |> arrange(-prop) |> head(4)

A tibble: 4 × 3

task	n	prop
<fct>	<int>	<dbl>
watch	3786	0.3460062
cell	3028	0.2767319
baseline	2736	0.2500457
voicecall	1392	0.1272162

A tibble: 4 × 3

id	n	prop
<fct>	<int>	<dbl>
16	455	0.04158289
4	431	0.03938951
27	417	0.03811003
23	406	0.03710473

• There is no evidence of class imbalance in any categorical variable in the dataset.

• For the task variable, the largest category watch accounts for about 34.6%, not a large majority of the observations.

• For the id variable, the largest category 16 accounts for about only 4.2%.

---

Task 3: Data visualization

8.

• The two numerical variables selected from the dataset are: eda and hr.
• A scatter plot is appropriate to explore the relationship between these two variables.

BRLS21_EDA |>
  ggplot(aes(x = eda, y = hr)) +
  geom_point(alpha = 0.3, size = 0.5) +
  ggtitle("Relationship between Electrodermal Activity Measure and Heart Rate") +
  xlab("Electrodermal Activity Measure") +
  ylab("Heart Rate")

Warning message:
“Removed 739 rows containing missing values or values outside the scale range
(`geom_point()`).”

• The scatter plot reveals that data points are widely spread, indicating no clearly linear relationship overall.

• Most observations cluster within the lower EDA range.

• As EDA values increase, data points gradually become more thinly spread.

• The HR variation range for the same EDA is relatively large.

---

9.

BRLS21_EDA |>
  ggplot(aes(x = eda)) +
  geom_histogram(fill = "steelblue", color = "white") +
  ggtitle("Distribution of Electrodermal Activity Measure") +
  xlab("Electrodermal Activity Measure") +
  ylab("Count")

`stat_bin()` using `bins = 30`. Pick better value `binwidth`.
Warning message:
“Removed 214 rows containing non-finite outside the scale range (`stat_bin()`).”

• The histogram reveals that the EDA distribution exhibits a right skew.

• Most observations cluster within the low to medium EDA range, while only a small number of observations exist in the higher value region, forming a long right tail.

• Based on this distribution, a typical EDA observation generally falls within the range of approximately 5 to 15.

---

10.

BRLS21_EDA |>
  ggplot(aes(x = task, y = eda, fill = task)) +
  geom_boxplot() +
  ggtitle("Electrodermal Activity Measure by Task") +
  xlab("Task") +
  ylab("Electrodermal Activity Measure")

Warning message:
“Removed 214 rows containing non-finite outside the scale range
(`stat_boxplot()`).”

• The box plots reveal noticeable differences in EDA distribution across different tasks.

• Under the baseline condition, the overall EDA level is the lowest, with both the median and range of distribution being smaller.

• The median EDA values under the cell, voicecall, and watch conditions are all higher, and the distribution ranges are broader.

• Specifically, higher EDA values appear more frequently under the voicecall condition.

• Overall, this chart indicates an association between task type and the level of EDA.

---

Task 4: Find summary statistics

12.

BRLS21_EDA |> summarise(median_eda = median(eda, na.rm = TRUE))

A tibble: 1 × 1

median_eda
<dbl>
9.366185

• The distribution of EDA has a right skew and contains a significant number of large numbers.

• Compared to the mean, the median is less impacted by the influence of extreme values and better represents the typical level of EDA.

• Therefore, the median is an appropriate measure for describing the central tendency of this numerical variable.

• The median of EDA is approximately 9.37.

---

BRLS21_EDA |> count(task) |> mutate(proportion = n / sum(n))

A tibble: 4 × 3

task	n	proportion
<fct>	<int>	<dbl>
baseline	2736	0.2500457
cell	3028	0.2767319
voicecall	1392	0.1272162
watch	3786	0.3460062

---

14.

BRLS21_EDA |> select(eda, hr) |> corrr::correlate()

Correlation computed with
• Method: 'pearson'
• Missing treated using: 'pairwise.complete.obs'

A cor_df: 2 × 3

term	eda	hr
<chr>	<dbl>	<dbl>
eda	NA	0.1435454
hr	0.1435454	NA

• The Pearson correlation coefficient between EDA and HR is approximately 0.14, indicating a weak positive correlation between the two.

• As the level of EDA increases, there is a slight overall upward trend in HR, but this association is relatively weak.

• There is no clear linear relationship between the EDA and HR.

---

Task 5: Develop a question

15.

Compared to phone notifications (cell), do smartwatch notifications (watch) results in higher levels of EDA while driving?

---

16.

• There is no need to split the data into training and testing sets.

• The objective of this research question is to compare the level of EDA across two kinds of task, not to train a new predictive model.

• This is a descriptive question, not a predictive question.

---

17.

answer_17 <- BRLS21_EDA |>
  filter(task == "watch" | task == "cell") |>
  group_by(id, task) |>
  summarise(mean_eda = mean(eda, na.rm = TRUE))

`summarise()` has grouped output by 'id'. You can override using the `.groups`
argument.

• The new feature derived from the existing variables is the average EDA for each participant (ID) under the watch and cell task conditions.

• First, filter observations where the task type is either watch or cell.

• Then, group by ID and task, and calculate the mean EDA within each group.

• Comparing the mean EDA of the same participant under watch and cell conditions helps more clearly evaluate the relative impact of the two notification tasks on EDA, making this a valuable feature.

---

Milestone 2: Sampling and Study Design

Task 6: Describe design

18.

• The type of data analysis questions the authors of the paper hoping to answer is inferential.

• The study proposes several clear hypotheses and uses hypothesis testing, reporting p-values to assess the correctness of hypotheses.

---

19.

• The Population used in this study: All drivers who receive and process notifications while driving.

• The Sample used in this study: 31 voluntary registered drivers between the ages of 18 and 47 who drive in a driving simulator while receiving and answering text messages.

---

20.

• This is a controlled experiment study.

• In the Section 3.1. Experimental Design of the paper, the researchers clearly states: "A within-subject experiment was conducted with the notification medium as the main independent variable."

• This indicates that the researchers actively controlled the independent variable and compared results under controlled conditions.

---

Task 7: Dive into the design

21.

• Researchers controlled for many confounding variables in this study.

• The same participant completed across all four kinds of tasks to reduce participant-related bias.

• The order of the four tasks was randomized to reduce order-related bias.

• Notification content was randomly assigned across the four tasks, with each task receiving 6 notifications to reduce notification-related bias.

---

23.

• There were several potential sources of sampling bias that may have affected the study’s conclusions.

• The age structure of the sample is skewed toward younger individuals as the author stated that "Twenty-four (24) out of 31 participants were aged between 20 and 26 years old."

• The smartwatch usage experience of the sample is relatively scarce in the sample as the author stated that "Only 9 out of the 31 participants were smartwatch users, with only one of them who used it for over two years, according to the questionnaire’s results."

• Participants were recruited voluntarily through the research panel of the researchers' institution. This may lead to the sample of drivers differing from the population of drivers.

---

Task 8: Now it is your turn

24.

• If I were to design this study myself, I would use a different sampling and design method. These improvements primarily address the shortcomings acknowledged by the author.

• Sampling:

  • Expand the age range of the sample.

  • Increase the proportion of smartwatch users.

  • Using smart devices of various brands and models.

  • Such changes make the sample more representative of the population.

• Design:

  • The author notes that due to the battery limitations of the eye-tracking glasses, there was no control condition. Perhaps a control condition could be added by using more advanced eye-tracking glasses or by shortening the test duration for each condition.

  • The author suggests that "More driving measures could be added in a future study for a better overall view of driving behaviour." Additional driving indicators should be incorporated as recommended by the author.

  • The author suggests that "screening for issues related to eyesight and vision should be a part of subsequent research." An eyesight check should be incorporated as recommended by the author.

• The authors might have chosen not to use alternative sampling or design methods due to:

  • There exist critical technical constraints, such as the battery temperature of tracking glasses.

  • The authors may have prioritized experimental efficiency over attempts to enhance sample representativeness. So they use participants who enrolled voluntarily.

• The potential limitations or challenges of my proposed approach are:

  • Technical constraints may not be resolved in the short term.

  • Recruiting more representative samples would be more difficult and costly.

  • Using multiple brands and models of smart devices may introduce new biases.

---